Ctffind4 - discrepancies between raw frames and realigned images

Submitted by celia on Wed, 2017-03-29 08:59

Forums

ctffind & ctftilt

Dear Niko, dear Alexis and all,

I have stacks of 7 frames recorded on a Cs-corrected Titan Krios with a Falcon II camera ; the calibrated pixel size is 1,12 A ; electron dose is 6.5 e/A2 per frame. I have noticed significant discrepancies in ctffind4 results when comparing results obtained either from « raw »frames (Defocus ~0.6 µm) or from re-aligned and re-stacked images (with unblur, no dose filter; Defocus : ~ 2.5µm). I am also puzzled by the power spectra (see attached image): I don’t understand why they look so different, aren’t they supposed to be both calculated from the FT of the whole image (and thus be quasi identical?)

I am very probably missing something here, but I don’t know what… Could you help me understand, please ?

many thanks in advance for your help,

Celia

Here is what I did in details :

1) took one stack of frames, used it as input in ctffind4 without averaging (or averaging 2, 3 or 7 frames)
⇒ got the following results :

Output from CTFFind version 4.1.5, run on 2017-03-22 12:00:24
# Input file: FoilHole_29178617_Data_29171393_29171394_20160229_1803_frames.mrc ; Number of micrographs: 1
# Pixel size: 1.119 Angstroms ; acceleration voltage: 300.0 keV ; spherical aberration: 0.00 mm ; amplitude contrast: 0.07
# Box size: 1024 pixels ; min. res.: 35.0 Angstroms ; max. res.: 5.0 Angstroms ; min. def.: 5000.0 um; max. def. 50000.0 um
# Columns: #1 - micrograph number; #2 - defocus 1 [Angstroms]; #3 - defocus 2; #4 - azimuth of astigmatism; #5 - additional phase shift [radians]; #6 - cross correlation; #7 - spacing (in Angstroms) up to which CTF rings were fit successfully
1.000000 6528.936523 6306.740234 56.677429 0.000000 0.035601 3.963576

I obtained similar values when averaging 2, 3 or 7 frames together.

2) re-aligned this stack with unblur (no filter dose), used the aligned sum as input in ctffind4. Using same search parameters as before, here is what I got :

Df1 : 25991.7 Df2 : 24862.9 (Of note, I obtained similar defocus values after re-aligning frames with MotionCorr).

# Output from CTFFind version 4.1.5, run on 2017-03-29 13:27:59
# Input file: FoilHole_29178617_Data_29171393_29171394_20160229_1803_frames_unblur_noF.mrc ; Number of micrographs: 1
# Pixel size: 1.119 Angstroms ; acceleration voltage: 300.0 keV ; spherical aberration: 0.00 mm ; amplitude contrast: 0.10
# Box size: 1024 pixels ; min. res.: 35.0 Angstroms ; max. res.: 5.0 Angstroms ; min. def.: 5000.0 um; max. def. 50000.0 um
# Columns: #1 - micrograph number; #2 - defocus 1 [Angstroms]; #3 - defocus 2; #4 - azimuth of astigmatism; #5 - additional phase shift [radians]; #6 - cross correlation; #7 - spacing (in Angstroms) up to which CTF rings were fit successfully
1.000000 25991.724609 24862.939453 55.689486 0.000000 0.021311 8.911890

Alexis Wed, 2017-03-29 10:13

Dear Celia,

At first glance, this looks like a pixel-size error. Is it possible that one of the files was resampled/binned, and the other not? Can you double check they have the same x-y dimensions?

You are correct - the two power spectra should look very similar, with the zeros in the same places. Looks to me like the first zero in Thon rings from the frames is ~60 pixels from the center, but in Thon rings from the aligned sum it is ~30 pixels from the center. Is it possible that the frames were resampled/binned 2-fold after they were given to unblur, but before they were given to ctffind?

Alexis

celia Wed, 2017-03-29 11:08

Dear Alexis,
thanks for your quick reply.

I checked (by displaying with imageJ) and both the stack of frames and the realigned image have the same dimensions (4096x4096). the initial stack of frames is 16-bits, and the unblurred image is 32, could that be an issue? Also I did not resample the frames stack or the unblurred image - at least not knowingly! I'm a bit lost here.

Celia

celia Wed, 2017-03-29 12:06

Hi Alexis,

I checked again and noticed something maybe weird in the header of my initial stack of frames (when read by ctffind4).
Here is what I did:

A) ran ctffind 4 on one of my initial, unaligned stack of frames (FoilHole_xxx_1742_stackall.mrc). below is a part of the output of ctffind4:

Summary information for file FoilHole_xxx_1742_stackall.mrc
Number of columns, rows, sections: 4096, 4096, 7
MRC data mode: 6
Bit depth: 16
Pixel size: 1.000 1.000 1.000
Bytes in symmetry header: 0
(...)
Estimated defocus values : 5000.00 , 4911.53 Angstroms

B) ran Unblur (no dose filter) on this stack of frames, asking two outputs: (1) sum after re-aligning ("FoilHole_xxx_1742_unblur_noF.mrc"),
and (2) realigned frames ("my-aligned_frames.mrc")

C) ran ctffind4 on both, this time got similar results between (1) and (2), but very different from initial stack

Ctffind4 on FoilHole_xxx_1742_unblur_noF.mrc (sum of realigned frames):

Summary information for file FoilHole_xxx_unblur_noF.mrc
Number of columns, rows, sections: 4096, 4096, 1
MRC data mode: 2
Bit depth: 32
Pixel size: 0.000 0.000 1.000
Bytes in symmetry header: 0
(...)
Estimated defocus values : 19706.13 , 19449.48 Angstroms

Ctffind4 on my-aligned_frames.mrc (realigned frames, not summed ; averaged 7 frames)

Summary information for file my_aligned_frames.mrc
Number of columns, rows, sections: 4096, 4096, 7
MRC data mode: 2
Bit depth: 32
Pixel size: 0.000 0.000 0.143
Bytes in symmetry header: 0
(...)
Estimated defocus values : 19706.39 , 19449.48 Angstroms

So I wonder, could this be some kind of bug in the header of the initial stack of frames?

Thanks again for your help,
Celia